[Bugfix] Set PREEMPTED status when moving requests from running to waiting queue

[gcanlin]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

When the running queue exceeds scheduler_max_num_seqs in async chunk mode, requests are moved to the waiting queue. The base scheduler expects waiting queue requests to have WAITING or PREEMPTED status, but the requests retained their RUNNING status, causing Invalid request status: RUNNING error.

Test Plan

Qwen3-Omni:

vllm bench serve --omni --dataset-name random --port 8091 --max-concurrency 10 --model Qwen/Qwen3-Omni-30B-A3B-Instruct --endpoint /v1/chat/completions --backend openai-chat-omni --num-prompts 100 --random-input-len 100 --ignore-eos --percentile-metrics ttft,tpot,itl,e2el,audio_ttfp,audio_rtf --random-output-len 100 --extra_body '{"modalities": ["text", "audio"]}'

vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /root/vllm-workspace/vllm-omni/vllm_omni/platforms/npu/stage_configs/qwen3_omni_moe_async_chunk.yaml

Test Result

I tested on NPU:

============ Serving Benchmark Result ============
Successful requests:                     100
Failed requests:                         0
Maximum request concurrency:             10
Benchmark duration (s):                  868.59
Request throughput (req/s):              0.12
Peak concurrent requests:                14.00
----------------End-to-end Latency----------------
Mean E2EL (ms):                          81511.37
Median E2EL (ms):                        84425.49
P99 E2EL (ms):                           134560.01
================== Text Result ===================
Total input tokens:                      10000
Total generated tokens:                  10000
Output token throughput (tok/s):         11.51
Peak output token throughput (tok/s):    100.00
Peak concurrent requests:                14.00
Total Token throughput (tok/s):          23.03
---------------Time to First Token----------------
Mean TTFT (ms):                          530.21
Median TTFT (ms):                        425.05
P99 TTFT (ms):                           1576.20
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          69.72
Median TPOT (ms):                        63.12
P99 TPOT (ms):                           105.19
---------------Inter-token Latency----------------
Mean ITL (ms):                           69.02
Median ITL (ms):                         60.75
P99 ITL (ms):                            505.09
================== Audio Result ==================
Total audio duration generated(s):       2517.04
Total audio frames generated:            60409080
Audio throughput(audio duration/s):      2.90
---------------Time to First Packet---------------
Mean AUDIO_TTFP (ms):                    10181.04
Median AUDIO_TTFP (ms):                  7630.57
P99 AUDIO_TTFP (ms):                     33740.78
-----------------Real Time Factor-----------------
Mean AUDIO_RTF:                          3.24
Median AUDIO_RTF:                        3.23
P99 AUDIO_RTF:                           3.36
==================================================

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[gcanlin]

cc @amy-why-3459 @linyueqian @Sy0307

[hsliuustc0106]

do we have any plan the speed up NPU perf?

[gcanlin]

do we have any plan the speed up NPU perf?

To be honest, currently only one or two are working on Omni models. I’m not sure whether we’ll make significant progress in March, but I’ll do my best to optimize them.

[amy-why-3459]

LGTM

[hsliuustc0106]

lgtm